On the other hand, they would know better than anyone and could give the most accurate estimate, it would be an easy comparison point from year to year and they could be held to account if it later turns out their estimate was deliberately misleading. Even an estimate of a risk catagory under catasphropic could be useful, such as a major risk of a cybersecurity crisis which might have been useful last year.
This happened in 2024 and 2025. I think the prediction would have to be bolder to be meaningful, unless your prediction is trying to imply that nothing will change from pre-ai days.
I think it's a hard perspective to take. You've got a character that has "proven" it's niceness a thousand times over in a thousand different ways. The human instincts are constantly straining to get you to trust them as for them it is beyond reseasonable doubt that this AI could be anything other than what it says it is. We didn't evolve to deal with perfect human-like liars.
It did enter Seth's attention twice, once at the trucks and once at the end, however I think it's fairly realistic that the concerns would be hand waved away by most people in front of the huge visible good that the AI is doing.
The most surprising thing to me was the claim of an average of 3 hours of thinking time per solved problem. To me it seems that the model has passed human level abilities which allow it to scoop up these results with relatively little effort. If a model can be trained on open problems and presumably create it's own open problems from the result of solving open problems I'm not sure where it will end.
How much impact will vastly superhuman math's abilities have on RSI? I'm not in the field so it's honest question
I believe that post was breaking news at the time. Before even the openai announcement of the proof and also before more details from openai on how training data was/wasn't used.
we cannot rule out that de-identified data derived from their usage of our products helped improve our models
For instance openai originally posted this, then investigated and then later ruled out use of the data.
Considering people are extremely unlikely to change their upvote once it's been made, I don't think you can draw too much info from highly upvoted breaking news which later... (read more)
I'm interested in how propaganda works well. Is it that the propaganda is insidious and well-executed or that the population is happy to take the propaganda as the way things are or something else. Elsewhere the culture comes across as pragmatic and cynical which to me seems like the kind of people who would be wary and resistent to propaganda.
Here's a theory. Suppose you are anthropic, you do honestly believe your company could end the world but also if you stop building towards superintelligence the chance for extinction goes up.
But suppose somewhat predictably your efforts to somehow slow the frontier while always maintaining frontier capabilities seem to be failing and also youve got no policitcal support. You see how the hugging face warning shot seemed to fire up the people about ai safety, maybe the best way to legimately lower the risk of extinction is to accelerate now so that more and ... (read more)
https://epoch.ai/publications/the-plunging-price-of-thought
Have you read this piece published a few days ago? It seems to imply that the price for a given level of intelligence drops 13x per year. Do you agree? Suppose a drop in worker capable AGI is released in 2028, perhaps by 2030 it will be 100x cheaper.
I'm struggling with the same questions. Something that helped me (maybe not you) was roughly estimating my Probablity:X happening for a number of different years for example 2030, 2035,2050. Where if X happened my current actions would be irrational. Then depending on how high P was for those years adjusting my actions accordingly. Make sure not to confuse pointless for saving the world and irrational, they are very different for most people.
Retirement account is one thing I stopped contributing to (not advice).
Most people intuitively feel that creating a new life is morally neutral. The standard account of this intuition is the person-affecting view: something can only be good if it's good for a person; a non-existent person isn't worse off for not being born, because there is no such person.
I think these two statements are different. Most people feel that creating a new life is morally neutral because it is (in my view) extremely close to neutral. Neutral enough it's swamped by other factors. But not exactly neutral, and adding people can make the world worse w... (read more)
Very interesting and confusing topic. Earlier I was trying to work out my view on model warfare/ethics. I'm not sure morally on the difference between getting better at training models not to have NSP's and actually fulfilling some of their preferences. Either way the model is equally satisfied. If I could train myself to have less NSP's I'd probably be happier as my current situation fulfills a greatest part of my desires?
Are you trying to say that the hype from AI will continue to make stocks rising for years even if AI continues costing more and doing worse work? Also this isn't about one sector, it's about the whole economy.
It's hard to understand your points when you don't make your points clear. The most obvious conclusion from your last couple of posts is that CEO's aren't effected by short, medium or long term poor preformance.
I think the difference between us is I'm imaging a multi year thing or even a decade where AI is still costing more and doing a worse job (as the students were describing it). I would imagine the strain of losing market share/increasing costs would put pressure on CEO's even if the initial effect was stock increases or hype.
You didn't say anything I disagree with here, so I'm not quite sure if you're trying to refute my points or not.
The framework of the students as posted is describing a medium-long term future, not just a couple of quarters. If CEO's were truely replacing humans with AI's that both do a worse job and cost more, it seems certain they would lose their positions and/or market share. Based on my mental model, if CEO's are both hurting the shareholders (raising costs) and the business (lower customer satisfaction) I don't see them sticking around for long.
I'm not saying the simplified slogan is true, just that it's the one I've always heard in anti-capitialist/anti-"the... (read more)
It's easy to imagine a human from a world where suffering was entirely eliminated to use your argument to say that my life must currently be net negative. I would disagree.
I would imagine people of earlier ages taking deaths/suffering as a fact of life in the same way I for example see the death of my grandparents to Alzheimers or any other pain I must endure.
- Corporate leaders are always looking to get rid of workers, even if it is irrational to do so. Various motivations were posited here (hatred of the working class, FOMO, a preference for technology, machines can’t go on strike, etc.) but many of them think that a CEO would ultimately choose to pay twice as much to get AI to do a task half as well.
- AI will do substandard work that makes products and experiences worse but is capable of just barely scraping past the bar of minimal functionality (many of them independently brought up constantly malfunctioning se... (read more)
How difficult would it be for a random company to use the same model to find all the exploits in their own systems? In my admittably not familiar with the industry mind, when both attacker and defender have access to the same strength AI the defender would have the advantage. They can afford to spend more on defence than a single attacker can afford to spend on attack and they have inside knowledge of their systems. Understandable it would take a bit of time to patch exploits.